Track Clustering Using Fréchet Distance and Minimum Description Length

Abstract

Clustering of objects into similar groups that share some common attributes has been a problem of interest in the data-mining community. Applications of this sort include pattern recognition, data analysis, and image recognition. In one recent track data-mining application, a dataset of polygonal trajectories was analyzed to discover common subtrajectories. Recent approaches to the track data-mining problem have made use of either the principle of minimum description length or new metrics for computing the distance between objects with a polygonal shape. A new approach to the track clustering problem based on the Fréchet distance metric and the minimum description length principle is proposed and tested with the GeoLife dataset. This approach can be generalized for clustering any dataset of shapes on a metric space.